Researchers gaslit Claude into giving instructions to build explosives

Anthropic has spent years building itself up as the safe AI company. But new security research shared with The Verge suggests Claude's carefully crafted helpful personality may itself be a vulnerabili...